BI and data visualization
Choose for the people asking questions and the data they may access.
A SQL analyst exploring a warehouse, a business team building dashboards and a customer viewing an embedded report need different interfaces and permission models. I would evaluate those workflows before counting charts. The database still executes the analytical workload; BI adds its own metadata, identities, cached results and publishing responsibilities.
-
Datasets and metrics
Curated tables, semantic models or versioned SQL queries define dimensions, measures and joins so teams share consistent definitions.
-
Identity and permissions
SSO, role-based access, row-level filters and database credentials determine whether queries run as a shared service role or the viewer.
-
Query pushdown and cache
The BI layer compiles visual explorations or SQL into warehouse queries, manages async workers and caches results without leaking tenant data.
-
Dashboards, reports and embeds
Interactive charts, scheduled deliveries, Markdown-as-code reports and signed embeds reach internal readers or external customers.
At a glance
Sources reviewed: October 2026, from official repositories, releases and documentation. Each rating (1–5) is an editorial readiness score for the fit in its row; the breakdown sits under the table and the method on the Blog index.
| Project | Rating | Core license | I would shortlist it for | Main constraint |
|---|---|---|---|---|
| Apache Superset 6.1.0 | Apache-2.0 | SQL exploration and shared dashboards over several analytical engines | Metadata, workers, caches and database permissions need an operating plan. | |
| Metabase 0.64.1 | AGPLv3 for Open Source Edition | Business users combining a visual query builder with SQL questions | Advanced governance and embedding capabilities depend on the edition and terms. | |
| Redash 26.9.0 | BSD-2-Clause | SQL-led teams sharing queries, charts and dashboards | Scheduling, workers, query results and source credentials need maintenance. | |
| Lightdash 2.586.0 | MIT core; enterprise exceptions | Exploration built around modeled dimensions and metrics, especially with dbt | Model ownership and the exact open-core feature set must fit the deployment. | |
| Evidence 40.1.8 | MIT framework | Authored data reports using SQL and Markdown | Publishing and access control must protect the data shipped with the report. |
How each rating was computed
Five criteria worth 0, 0.5 or 1 point each. Caps: archived upstream 1, no stable release in 18 months 2, no general-availability release 2.5. "Open edition" scores what the open-source edition includes without a paid tier. Scored in October 2026 from the official repository, releases and documentation; the method is on the Blog index.
| Project | Maintenance | Open edition | Maturity | Operations | Interoperability | Rating |
|---|---|---|---|---|---|---|
| Apache Superset | 1 | 1 | 1 | 0.5 | 1 | 4.5 |
| Cube | 1 | 1 | 1 | 0.5 | 1 | 4.5 |
| Metabase | 1 | 0.5 | 1 | 0.5 | 1 | 4 |
| Redash | 0.5 | 1 | 0.5 | 0.5 | 1 | 3.5 |
| Lightdash | 1 | 0.5 | 0.5 | 0.5 | 0.5 | 3 |
| Evidence | 0.5 | 0.5 | 0.5 | 1 | 0.5 | 3 |
Start with the reader
- SQL analysts exploring multiple sources: shortlist Superset and Redash. Test the real database drivers, query limits and dashboard workload.
- Business users who need guided exploration: start with Metabase and trial a real question with the intended users. Verify the controls in the Open Source Edition before including paid features in the design.
- A team already governing metrics with dbt: compare Lightdash against the existing semantic definitions and ownership process. Reconcile joins, filters and metric names before choosing the presentation layer.
- A report with a curated explanation and reproducible publication: consider Evidence. Treat the generated data assets as part of the published artifact, with an explicit audience and access boundary.
- Analytics embedded in a customer application: evaluate tenant isolation, identity propagation and licensing first. An iframe alone does not establish an authorization boundary.
Version and operating notes
Superset — protect both the application and the database
The reviewed repository describes visual exploration, SQL Lab and database connectivity. Its security guidance explains permissions and row-level restrictions.
Use a database identity with only the required privileges. Test cached results, SQL access and exports with two users who should see different data. Back up the application metadata as well as the analytical source; dashboards and permissions are not reconstructed from warehouse tables.
Metabase — identify the edition and embedding terms
The license file distinguishes AGPL code and binaries from enterprise artifacts. The licensing page documents separate embedding options and commercial terms.
Trial the question builder, SQL editor and sharing workflow with the intended audience. Check the exact edition for tenant isolation, audit information and identity integration. Recover the application database and verify saved questions, permissions and scheduled deliveries.
Lightdash — keep metrics and application changes coordinated
The project connects governed metrics and dimensions to interactive exploration. Its license makes the core MIT, with an enterprise directory under separate terms.
Change a metric definition and a join while existing dashboards still reference them. Check review, preview and rollback in the actual deployment. Validate the dbt integration and supported adapters without assuming every feature in the hosted product ships in the core.
Evidence — review the published data artifact
Evidence Core builds data applications from SQL and Markdown. Its MIT framework and hosted Studio have distinct feature scopes; Studio-specific capabilities are identified in the docs.
Inspect the built artifact and browser requests with a reader account. Hiding a chart or navigation item does not remove its underlying data. Trial updates, failure during publication and recovery of the previous report. Verify the documentation matches the chosen framework generation.
Redash — operate the query execution path
The self-hosted project supports saved queries, visualizations and dashboards. The documentation separates user workflows from operating a self-hosted instance.
Test a long-running query, unavailable data source and expired credential. Check which results survive in caches and who can export them. Budget the application database, queue and workers, not only the web process.
Where Cube and Grafana fit
Cube 1.8.6 is a semantic layer for exposing modeled data through APIs and SQL. Its code licensing combines Apache-2.0 and MIT by package. It can support an analytics application; it is not the same deliverable as a complete self-service BI interface.
Grafana belongs primarily in the observability comparison here. It can query SQL sources, but incident investigation, time-series dashboards and business metric governance are different evaluation scenarios.
Candidates to watch
Rill (Apache-2.0; 0.90.5, released 2026-10-09) defines dashboards, a semantic layer and security policies as YAML and SQL, querying OLAP engines such as ClickHouse and DuckDB. It is not rated yet: I have not reviewed what the open-source edition includes against Rill Cloud under the criteria above.
A trial that can change the decision
Use one revenue model, two tenants and a representative business question.
- Correctness: reconcile totals, time zones, nulls, filters and join fan-out with a reference query.
- Authorization: verify dashboards, direct query access, exports, cached results and embedded views after revoking a user's permission.
- Usability: observe intended users completing the question; record errors and assistance required, not only elapsed time.
- Recovery: restore metadata, identities, connections and reports, then confirm scheduled delivery reaches the correct audience.
- Cost: measure source-database work, refresh frequency, cache storage and operating effort with representative concurrency.
Follow the data
Transformation defines trusted models; OLAP engines execute analytical queries. The table format governs lakehouse state, while observability helps detect stale reports and failures. Start with the smallest combination that meets the reader's requirements.